The mechanisms underlying synaptic transmission at the layer 4 of sensory cortical areas

The mechanisms underlying synaptic transmission at the layer 4 of sensory

cortical areas

PhD Thesis

in partial fulfillment of the requirements for the degree ‗Doctor of Philosophy (Ph.D.)‘

in the Neuroscience Program

at the Georg-August-University Göttingen, Faculty of Biology

submitted by Chao-Hua Huang

born in

Kaohsiung, Taiwan

2010

Declaration

I hereby declare that my PhD thesis ‘The mechanisms underlying synaptic transmission at the layer 4 of sensory cortical areas’ has been written independently with no other aids or sources than quoted.

Göttingen, September 20th 2010 ………..

Contents

1. Introduction ... 1

1.1 Motivation ... 1

1.2 Top-down, system approaches to the visual function ... 1

1.3 Bottom-up, cellular approaches to the visual function... 3

1.4 From a cell to a synapse ... 4

1.4.1 Fluctuation analysis ... 4

1.4.2 ―one-site, one-vesicle‖ hypothesis v.s. multivesicular release... 8

1.5 Aim of this study ... 10

2. Material and Methods ... 11

2.1 Slice preparation ... 11

2.2 Electrophysiology ... 14

2.3 Variance-mean analysis ... 16

2.4 Deconvolution ... 18

2.5 Dynamic clamp experiments ... 19

3. Results... 20

3.1 Variance-mean analysis and the mEPSC amplitude measurement at L4 neurons in V1... 20

3.2 Estimation of N, Pr, and q under physiological condition with variance-mean analysis ... 25

3.3 Depletion of the readily releasable pool and calculating release rates with the deconvolution method ... 27

3.4 Saturation of postsynaptic receptors at RS-RS connections in S1 ... 29

3.5 A close look at the blocking effect of Kyn suggests MVR in S1 ... 34

4. Discussion ... 49

4.1 Multivesicular release at cortical synapses... 49

4.2 Differences of synapses at two sensory areas and their physiological implication . 51 4.2.1 Differences in Pr and replenish rate ... 51

4.2.2 Difference in reliability of synaptic transmission revealed by dynamic clamp 52 4.2.3 Summary of physiological and anatomical data ... 55

5. Summary ... 58

Bibliography ... 59

A. Appendix ... 69

A.1 Abbrievations ... 69

Acknowledgement ... 70

Curriculum Vitae ... 72

Publication list ... 73

Introduction

1.1 Motivation

―Do we perceive the world passively or do we create the world actively by our brain, and how?‖ Sensory system plays a key role in our lives; indeed we ―define‖ the world according to how we ―sense‖ it. To understand how sensory systems function hence is a central issue in neuroscience. Various approaches like psychophysics, anatomy, physiology, and molecular biology were taken to solve this issue. Although the question is so big that till now we still don not have a complete answer to it, thanks to all the efforts of neuroscientists, the general anatomical structure of different sensory pathways was established. Among all kinds of senses, the visual system is crucial for higher animals. With the delicate design of the visual neuronal network, we are able to perceive the beauty of the world; therefore I would like to understand the circuitry within the visual system. Much is known about the general neuroanatomy of the visual sensory system by now. It has been long established that the visual sensory system begins at the retina. Light is received here and then the information is transmitted through the optic nerve and thalamus and finally to the visual cortex.

1.2 Top-down, system approaches to the visual function

The visual cortex can be split up into different areas: the primary visual cortex (V1), V2 area, V3 area, V4 area, and V5 area, and they represent different stages of visual information processing (Kuffler S. et al., 1984). Among them, the best studied area is V1

is essential for the most basic visual information processing. In V1, information first arrives in layer 4 (L4), then goes on to layer 2/3 (L2/3), and finally is sent out to the higher cortex or thalamus through layer 5 (L5) or 6 (Ferster and Lindstrom, 1983). In addition to the anatomical findings, a special feature of the sensory neurons in the visual pathway called a receptive field has been found physiologically. Extensive research has defined the receptive field of a neuron, which is a spatial (and temporal in some cases) pattern in which the stimuli excite or inhibit the firing of the neuron, and this defines the selectivity of the neurons to particular sensory inputs. The receptive fields can be different in size and shape. In the retinal ganglion cells and the thalamic relay neurons, the receptive field is circular with a concentric On/Off center/surrounding whereas cortical neurons have elongated receptive field, which has a specific orientation preference (Hubel and Wiesel, 1962). The neurons with certain elongated receptive field respond the best to a bar with special orientation instead of a spot in the receptive field like the neurons of the retinal ganglion and the thalamic relay neurons.

The differences in the receptive field properties have been an area of extensive research. It has been hypothesized that the differences in the receptive field properties reflect the processing of visual information. More importantly, it is postulated that the orientation selectivity makes us be able to ―see‖ the outline of an object which is the precondition for V1 to perform one of its tasks, detecting luminance contrast (see above).

Furthermore, a distinctive structure, the orientation selective column, has been identified in V1. This column intersects through the six-layer structure of the cortex, and the neurons in each column share the receptive field with similar orientation selectivity. The signal of the preferred orientation is transmitted mainly within the column (Hubel and Wiesel, 1962) which indicates that a well-defined circuitry exists to perform a specialized function which is the orientation selectivity in V1.

Various models have been suggested how the orientation selectivity emerged. Hubel and Wiesel (Hubel and Wiesel) postulated that orientation selectivity emerges in the thalamocortical synapses. They argued that the convergence of thalamo-cortical synapses to a cortical neuron is sufficient to form an elongated receptive field. However

in contrast to the Hubert and Wiesel model, Creutzfeldt and colleagues (Pei et al., 1994) have proposed that the intra-cortical connection plays a major role in orientation selectivity, although the importance of the thalamocortical synapses cannot be neglected.

These models have been around for over 40 years, but there is still no general consensus on how orientation selectivity emerges. Also a detailed explanation of the network on cellular level is lacking due to the resolution of the techniques such as extracellular recordings in vivo and functional magnetic resonance imaging which were used in previous researches. Nevertheless, despite the differences in the models it has been agreed that, L4 neurons, as the first gate for signals to get into the cortex, play an important role in filtering the visual signal and the formation of elongated receptive field.

1.3 Bottom-up, cellular approaches to the visual function

To compensate for the top-down approach, neuroscientists have come from the other end of the spectrum—a bottom-up approach was taken. Anatomical classification of cell types and connection patterns has been characterized in cat visual cortex (Ahmed et al., 1994; Binzegger et al., 2004; Gilbert, 1983; Gilbert and Wiesel, 1979). Physiologically, the neurons in each layer of neocortex were classified according to their firing patterns upon a current injection. Majorly three classes were described. Regular spiking neurons (RS) show adapting trains of spikes, and each spike has a relatively long half-width with complex afterhyperpolarization and afterdepolarization. Fast-spiking neurons (FS) fire at a high frequency with no adaptation, and the spikes they generate are brief with fast, deep monophasic afterhyperpolarization. The third type is intrinsically bursting neuron. This type of neurons has a tendency to generate a burst responding to a just-threshold stimulus.

Despite the burst, an individual spike of intrinsically bursting neurons is quite similar to that of RS neurons (Connors and Gutnick, 1990). Gupta A. et al. further expanded these three types into nine classes. Firstly, neurons were broadly classified as non- accommodating, accommodating, and stuttering cells. These three major classes were then subdivided into three sub-groups according to the characteristics of the onset of the response (Gupta et al., 2000). In principle, pyramidal cells/ spiny stellate cells belong to

other interneurons belong to the rest of firing types which form inhibitory synapses.

Furthermore, descriptive synaptic features such as short-term depression/facilitation were also documented (Beierlein et al., 2000, 2003; Gupta et al.). For example, RS to FS connection is a depressing synapse whereas RS to low-threshold spiking neurons (LTS, another subtype classified by Beierlein M. et al., 2000 which should belong to the accommodating class) is a facilitating synapse (Beierlein et al.). Moreover, molecular identity of different classes of neurons was also documented (Monyer and Markram, 2004). However, there is a huge gap between the knowledge at the cellular level and the functional system level. The logic of synaptic connections and the mechanism underlying the neural circuitry still remain unclear in visual cortex. If we take the whole visual pathway as a big machine which processes the raw materials (light stimulation) into the final product (images we perceive), knowing how the cables are wired (anatomy) and which possible computational units (cell type and descriptive features) exist is not enough to explain how the machine produce the product. Without knowing how each computational unit operates upon different inputs (synaptic transmission mechanism) and the current weight of each cables carries (synaptic strength and synaptic plasticity), it is impossible to deconvolve the whole producing process into the circuit diagram.

Therefore, research on the mechanism of synaptic transmission and synaptic plasticity is essential to fill the gap.

1.4 From a cell to a synapse

1.4.1 Fluctuation analysis

Synaptic plasticity is a collative term of all kinds of modifications in synaptic strength depending on neuronal activities. According to the time scale of such changes, it can be further categorized into long-term and short-term plasticity. Long-term plasticity refers to those modifications which last longer than tens of minutes to hours whereas short-term plasticity happens within the range of milliseconds (ms) to seconds (s). My research will focus on the latter part because the time scale of incoming sensory signals is similar to that of short-term plasticity (Cook et al., 2003; Zucker and Regehr, 2002).

More importantly, short-term plasticity makes a synapse play an active role in real-time information processing with a strong computational power (Abbott and Regehr, 2004;

Abbott et al., 1997).

Synaptic strength is determined by both pre- and post-synaptic factors. Del Castillo and Katz established the quantum theory by careful observations of synaptic responses at the neuromuscular junction synapse. They found that the amplitude histogram plot of end-plate potentials recorded from the muscle showed a poisson/binomial distribution with several peaks. More importantly, these peaks are separated with equal distance which indicates that a synaptic response composes of at least one quantum, and the distribution of quantum amplitude is a normal distribution with the same mean value. A larger synaptic response is a sum up of several quanta. This beautiful property of synaptic response resulted in the key concept of quantum theory: the synaptic strength is set by three parameters, the number of releasable units (N), release probability (Pr), and the quantal size (q). These parameters in turn determine the physical constrains of synaptic plasticity (Zucker, 1973). It is worthwhile to note that the definition of N at Katz‘s time was just an physically ambiguous term, ―quantal units‖. Nowadays its physical identity is still controversial. It can refer to the number of independent release sites, the number of releasable vesicles, or morphologically defined number of active zones. The Pr is the probability that synaptic vesicles can be released. The q represents the synaptic response evoked by only one vesicle (one quantum). According to the definitions, a synaptic response can be formulated as below:

Post-synaptic current=N * p * q

Following the quantum theory, a binomial model was applied to the amplitude histogram plot of the post-synaptic responses to estimate these parameters in different preparations (Buhl et al., 1997; Gulyas et al., 1993; Korn et al., 1981). It was reported that the number of quantal peaks in the histogram plot matches with the number of anatomical synaptic contacts, which led the authors to postulate the ―one site, one vesicle‖ concept (see below

the amplitude histogram plot displays clear peaks, and this condition can only be achieved when N and/or Pr is relatively small. Therefore, fluctuation analysis was introduced so that the analysis can be applied more generally (Redman S.J., 1990; Faber D.S. and Korn H., 1991; Silver R.A. et al., 1998; Clements J.D. and Silver R.A., 2000;

Scheuss V. and Neher E., 2001). Initial attempts of fluctuation analysis were done by analyzing the coefficient of variation (c.v.) of post-synaptic responses to distinguish the source of synaptic plasticity and to estimate the three quantal parameters (N, Pr and q, Faber D.S. and Korn H., 1991; Feldmeyer D. et al., 1999; Feldmeyer D. et al., 2002). For the latter purpose c.v. analysis was not ideal because one of the three synaptic parameters has to be known or assumed in order to obtain the other two. Silver et al. hence developed variance-mean (V-M) analysis (multiple probability fluctuation analysis) which does not need strong assumptions to estimate the N, Pr, and q from synaptic responses. The variance and mean are calculated from the fluctuation of synaptic responses responding to one action potential (AP). An essential feature of this method is that it explores the synaptic response fluctuations at different Pr (altering [Ca²⁺]), and because of multiple points in variance-mean plot it provides more information about the underlying synaptic mechanisms. Here, the N was defined as the number of independent release sites. Assuming the vesicle release follows a binomial model, the V-M plot of synaptic response fluctuations at different Pr settings displays a parabola relationship.

From the V-M plot, one can estimate the N, Pr, and q. In addition, because the V-M analysis is based on the binomial model which assumes q does not change with time or under different conditions , the relationship of V-M plot would be distorted if this assumption is violated,. This property potentially helps us to detect the changes in q (for example, the change at the post-synaptic site due to post-synaptic receptor desensitization or saturation). Scheuss V. and Neher E. further extended to apply V-M analysis to the synaptic responses during a train of action potentials (Scheuss V, and Neher E., 2001).

Instead changing the Pr by for example altering [Ca²⁺], this method samples from the dynamic Pr responding to a train of APs. This way the experimental protocol is simplified because there is no need of measuring under various conditions. Therefore it can be potentially used more widely (Biro et al., 2005; Scheuss et al.; Taschenberger et al., 2002).

The V-M analysis was used in different preparations to gain a mechanistic insight of synaptic plasticity (Silver et al., 1998; Scheuss V. et al. 2002; Taschenberger H. et al., 2002; Silver et al., 2003; Biro A. et al, 2005). However, not much has been done in the cortical synapses due to the complexity of neuronal types and the lack of detailed knowledge of fine cortical anatomy. Nevertheless, the most investigated cortical region is the somatosensory cortex (S1). It is also called Barrel cortex for the distinct feature of barrel shaped structures in L4, and the feature has in fact made S1 the most popular region for cortical/sensory research. Each whisker is represented somatotopically in the large-scale anatomical structure of a barrel in L4 of the neocortex (Woolsey and Van der Loos, 1970). The barrel structure is formed by clustered L4 neurons which receive inputs from the same whisker (Armstrong-James and Fox, 1987; Simons, 1978; Welker, 1976).

This structure can be recognized easily in the acute brain slice preparation under light microscopy (Agmon and Connors, 1991). These characters therefore allow us to relate the physiological function in vivo to the neuronal circuitry at the level of individual neurons and synaptic connections in vitro. In S1, the synaptic parameters have been estimated in L4-L4 excitatory connections using c.v. analysis (assuming N is the number of morphological synaptic contacts) (Feldmeyer et al., 1999) and L4-L2/3 excitatory connections by V-M analysis (Silver et al., 2003). The latter one is the first application of V-M analysis to the cortical neurons. It was reported that the number of functional release sites was equal to the number of morphological synapses in this type of connection. Furthermore, it was suggested that each synaptic contact released a single vesicle (independent of Pr) and that the intrinsic Pr was high, so this type of synapse must be tuned to response reliably to spatially distributed, timing-based signals. In addition to L4-L2/3 synapses, the synaptic transmission between L4-L4 neurons was also reported to be reliable (Feldmeyer et al., 1999). However, it is obvious that different sensory cortex may have different anatomical characters. In cat visual cortex, clear structure of orientation columns and ―blobs‖ can be seen, but in rat or mouse there is no defined column structure like Barrel cortex in the visual cortex. The logic of how neurons are wired must be different, but does this difference also exist at the synaptic level? What

unit that makes our brain able to perform different tasks? Taking the same type of synapse in S1 as a reference, we explored the mechanism of synaptic transmission in V1 connections by V-M analysis with train stimulation in our research and tried to get closer to the answers of these questions mentioned above.

1.4.2 “one-site, one-vesicle” hypothesis v.s. multivesicular release

Another line of mechanistic insight rooted in the quantum theory is the manner in which synaptic vesicles are released upon a single nerve impulse. As mentioned above, the definition of N was originally the number of releasable units which did not refer to any specific entity with a physical meaning. It could be the number of synaptic contacts, the number of readily-releasable synaptic vesicles, or the number of functional release sites. Based on different physical interpretations of N orgininally defined by Katz and his colleagues, two distinct hypotheses of the release process were proposed. The first hypothesis, so-called ―one-site, one-vesicle‖ suggests that only one vesicle can be released at one synapse upon one AP (Biro et al., 2005; Buhl et al., 1997; Egger et al., 1999; Feldmeyer et al., 1999; Gulyas et al., 1993; Korn et al., 1981; Murphy et al., 2004;

Silver et al., 2003). Korn H. et al. did simultaneous recordings and histological reconstruction on inhibitory synapses at the goldfish Mauthner cell. It was shown that there is only one active zone at one terminal bouton, and the binomial N matches with the number of bouton identified by the reconstruction; therefore the idea that the site number, N, refers to the number of anatomically defined synaptic contacts was proposed. More evidence of the binomial N matching with anatomical synapse number was reported to support the idea (Biro et al., 2005; Buhl et al., 1997; Egger et al., 1999; Feldmeyer et al., 1999; Gulyas et al., 1993; Murphy et al., 2004; Silver et al., 2003). Since the number of synapses follows the binomial model, only one vesicle can be released at one AP from one synapse. Based on this model, changes in synaptic strength at one synapse are all or none, and thus reduced the computation power of a single synapse. In addition, the rate of vesicular replenishment following a release event becomes crucial for a steady-state rate of neurotransmission (Dobrunz and Stevens, 1997; Stevens and Wang, 1995).

Furthermore, following this hypothesis, a mechanism explaining the ―single vesicular

constraint‖ was proposed by Dobrunz et al. in 1997. They observed that after one vesicle is released, the second one cannot be released for a short period (tens of ms), and they suggested that this is how only one vesicle can be released by one AP at one synapse.

On the other hand, an alternative hypothesis is multivesicular release (MVR) which provides more flexibility by allowing variation in the number of vesicles released by an AP (Li et al., 2009; Oertner et al., 2002; Tong and Jahr, 1994; Wadiche and Jahr, 2001;

Watanabe et al., 2005). Tong and Jahr first reported that the glutamate concentration in the synaptic cleft changes at different Pr. If there is only one vesicle released at one synapse, the glutamate concentration should be the same regardless of Pr. Therefore, they suggested that indeed one synapse can release more than one vesicle which results in a different interpretation of binomial N. The N refers to the number of functional release sites instead of the number of synaptic contacts. The phenomenon of change in glutamate concentration at synaptic cleft was also reported in several different preparations with different methods (Li et al., 2009; Oertner et al., 2002; Wadiche and Jahr, 2001; Watanabe et al., 2005). According to this hypothesis, a synapse can actually function as an analog device whereas the ―one-site, one-vesicle‖ hypothesis predicts a single synapse as a digital device, and as a result, multiple synapses are required to show graded modulation of signals. More computation power is given to a synapse with MVR.

In addition, MVR reduced the failure rate of vesicle release at one synapse at the same Pr.

Considering the condition of N=3 and Pr=0.6, MVR would predict a failure rate of (1- 0.6)3 = 0.064 at one synapse (assuming all three sites locate at one synapse) whereas the

―one-site, one-vesicle‖ hypothesis predicts a failure rate of (1-0.6) = 0.4 at one synapse.

Moreover, asynapse with MVR suffers more from the saturation/desensitization of post- synaptic receptors which further reduces the fluctuation of synaptic strength (Foster et al., 2002; Harrison and Jahr, 2003; Trussell et al., 1993). This does not happen with ―one- site, one-vesicle‖ hypothesis because no matter the post-synaptic receptors are saturated/desensitized or not, the c.v. is the same, and the fluctuations at a synapse level are all or none. Combining lower failure rates and modulations of post-synaptic receptors, MVR makes a synapse more reliable. Taken together, it is important to identify the

computational unit of a synapse and different prediction of synaptic reliability. As mentioned before, MVR was reported in several preparations such as hippocampal neurons (Oertner et al., 2002; Tong and Jahr, 1994), climbing fiber-Purkinji neuron synapses (Wadiche and Jahr, 2001), inner hair cell synapse (Li et al., 2009), and Calyx of Held (Taschenberger et al., 2002). However, the existence of MVR, particularly at small cortical synapses, remains controversial (Feldmeyer et al., 1999; Silver et al., 2003).

1.5 Aim of this study

In this study, we determined the quantal properties of excitatory connections betweenRS neurons in L4 of V1 and S1 quantitatively. This connection is known as a model system of cortical synaptic transmission as previously described (Feldmeyer et al., 1999; Petersen, 2002; Stern et al., 1992). These neurons in L4 receive signals from the thalamus and further transmit them to other layers in the cortex. At the same time, synaptic interactions between RS neurons often connect to each other within a short distance which allows better clamp for electrophysiological approaches (Lefort et al., 2009; Lubke et al., 2000; Petersen and Sakmann, 2000). In this study we estimated basic properties of small cortical synapses in two major cortical regions. By comparison of these properties, we addressed the issue of synaptic homo/heterogeneity in the cortex.

Furthermore, dynamic clamp experiments revealed the importance of such heterogeneity at synaptic level to the reliability of synaptic transmission.

Material & Methods

2.1 Slice preparation

Coronal slices (300 µm) were prepared from the visual cortex of P22-P28 NMRI mice with a vibrating microtome (VT1000/S1200S; Leica, Wezlar, Germany) (Fig. 1).

We started to collect slices at one slice prior to when the hippocampus showed up (approximately 1000µ m from the rear end of the cortex). In total 3-4 slices were collected for following recordings.

Figure 1 Coronal slice from mouse visual cortex

Region where slices were dissected is located between the two black solid lines of the brain (lower panel, left). 3-4 slices with the thickness of 300µm were collected, and a representative slice is shown here (lower panel, right). Visual cortex locates in between the two white dotted lines, and the view with simultaneous paired recording pipettes under the 4x objective is shown in the upper panel. Though the boundary of each layer is not necessarily clear, L4 neurons can be identified according to the cell morphology and the relative location to other layers (see below electrophysiology).

For somatosensory cortex, thalamocortical brain slices (300 µm) were prepared from P19-P24 mice according to Agmon A. and Connors B.W.,1991 (Fig. 2). Different from coronal sections of the visual cortex, a 55° angle relative to the mid-line of two hemispheres was taken to preserve the thalamocortical fibers. These fibers are the landmark of the start of collecting slices for recording. At least 6 slices could be

collected. If the cutting angle is correct, the barrel structure should be visualized clearly, and recordings were done on the RS neurons within barrels.

Figure 2 Thalamocortical slice from mouse somatosensory cortex (adapted from Agmon and Connors, 1991)

The cutting angle and the region where slices were dissected and collected are shown in the lower panel, left. 3-4 slices with the thickness of 300µm were collected, and a representative slice is shown here (lower panel, right). S1 cortex locates in between the two white dotted lines, and the barrel structure can be seen clearly even with naked eyes. The upper panel shows the enlarged view of the square region in the lower panel, left. Cells located in the barrels were chosen for our recordings.

The slicing solution contained the followings (in mM): 125 NaCl, 2.5 KCl, 25 NaHCO₃, 1.25 NaH₂PO₄, 3 MgCl₂, 0.1 CaCl₂, 0.4 L-ascorbic acid, 3 myo-inositol, 2 pyruvate, and 25 glucose. Before slicing, the slicing solution was frozen till half ice and half liquid, and to be bubbled with 95 % O₂ / 5 % CO₂. For animals older than P26 for visual cortex and for those older than P23 for somatosensory cortex, a sucrose solution

(in mM) 60 NaCl, 2.5 KCl, 120 sucrose, 25 NaHCO3, 1.25 NaH2PO4, 3 MgCl2, 0.1 CaCl2, 0.4 L-ascorbic acid, 3 myo-inositol, 2 pyruvate, and 25 glucose. After a brain slice was cut, it was transferred to a chamber with extracellular solution, and was incubated at 37˚C for an hour. During the incubation, the solution was always bubbled with 95%O₂/5%CO₂. The extracellular solution contained (in mM) 125 NaCl, 2.5 KCl, 25 NaHCO₃, 1.25 NaH2PO4, 1 MgCl2, 2 CaCl2, 0.4 L-ascorbic acid, 3 myo-inositol, 2 pyruvate, and 25 glucose. After incubation, the slices were stored in the same solution at room temperature and was bubbled with 95 % O₂ / 5 % CO₂.

2.2 Electrophysiology

L4 neurons can be recognized by the round shape and the small size of the somata (Feldmeyer et al., 1999; Stern et al., 1992). Also, the relative location of those neurons in the six-layer structure of the cortex was carefully examined. For visual cortex, there is no clear boundary in between each layer, but the neurons in different layers show distinct morphological characters, so one can still tell in which layer the recordings were done.

From the outer most edge of a slice, the molecular layer is firstly encountered. No cells located in this layer. With the direction to corpus callosum, the next layer to molecular layer is L2/3 where the neurons are pyramidal shaped with medium size surrounded by interneurons. And then the next layer is L4 where the neurons have round shape and relatively small size. L5 comes next to L4 and L5 neurons are distinct due to the large size and clear pyramidal shape. The last layer, layer 6, is characterized by the small round shaped neurons with lots of fibers running over the region. Normally, we looked for L5 neurons first because they are easy to be recognized. After finding L5, we moved toward the molecular layer until the small, round L4 neurons were seen. To further confirm they are indeed located in L4, we moved further toward the molecular layer until the characters of L2/3 neurons were found. L4 neurons should be located in between L2/3 and L5. Whole-cell voltage clamp recordings were performed on the soma of L4 neurons. After a connected pair was identified, excitatory post-synaptic currents (EPSCs) under different conditions were recorded (EPC 10, HEKA, Lambrecht, Germany) at a sampling rate of 50 kHz (filtered at 3-6 kHz). The resistance of glass pipettes we used

was 3-5 MOhm, and the intracellular solution contained (in mM) 140 potassium gluconate, 20 KCl, 10 HEPES, 5 ATP-Mg, 5 Phosphocreatine, 0.5 GTP, and 0.2 EGTA and was adjusted to pH 7.2 with KOH. The osmolarity of this solution was around 330 mOsm. Liquid junction potential (around 10 mV) was not corrected. Series resistances were all below 20 MOhm, and 20-50% compensation was used. The data were further off-line filtered with a low pass filter at 1 kHz before analysis. In some experiments, 50

M D(−)-2-amino-5-phosphonopentanoic acid (D-AP5, Tocris) was applied, but N- methyl-D-aspartic acid receptors (NMDARs) did not contribute to the peak EPSC amplitudes in our study (Fig.3). To prevent saturation of postsynaptic α-Amino-3- hydroxy-5-methyl-4-isoxazolepropionic acid receptors (AMPARs), 0.5 mM kynurenic acid (Kyn, Tocris) was added to extracellular solution in some experiments. In addition, in some experiments 10mM tetraethylammonium (TEA, Sigma) was applied to increase the releasing probability. In RRP depletion experiments, a half or two third of the potassium gluconate was replaced with cesium gluconate and 10mM TEA was also applied to block potassium channels, and to prevent repolarization. Also, D-AP5 was always applied in this experiment. All experiments (except experiments in Fig.5 and Fig.6) were performed at 30-35 ˚C.

Figure 3 The effect of DAP5

At S1, it has been reported that NMDA receptor-mediated EPSCs were prominent at negative potential (Feldmeyer et al., 1999; Stern et al., 1992). We examined this issue in our experimental condition (more matured animals and under physiological temperature) at L4 synapses in S1. Under voltage clamp, control traces (grey) and the traces under DAP-5 (black) were averaged over 10 times and are displayed in A. In B, the ratio of the EPSC amplitudes (DAP5/Ctl) is displayed from the 1^st to the 5^th EPSC during a stimulus train. The values range between 1 and 1.2, indicating that NMDA receptors do not contribute to the peak EPSC amplitudes in our experiments.

2.3 Variance-mean analysis

50 Hz train stimulations were applied to a pre-synaptic neuron and EPSCs were recorded from a post-synaptic one. The stimulus train was applied every 10 – 20 sec and was repeated more than 20 times. The data were excluded from analysis when rundown

of the EPSCs was noticed. The amplitude of each EPSC was calculated from the average of five data points around the peak subtracting the baseline of each peak. The baseline was the average over 50 points (1 ms) just before the onset of EPSCs. The stimulation was repeated for more than 20 times and the amplitude of each peak EPSC was taken for analysis. We averaged the EPSCs of each peak (the 1^st, 2^nd, 3^rd and so on) over all stimuli and the mean and the variance were obtained. These two parameters of each peak were plotted against each other. The estimation of N, Pr, and q was done according to Scheuss and Neher (2001). Definition of N is the number of functional release sites or vesicle number of readily releasable pool (RRP) (assuming P_occupancy = 1) which should be a fixed number. Pr refers to the combination of probability of vesicle occupancy at the slot and vesicular release probability, and q is the postsynaptic current amplitudes induced by a single vesicle release (Verejone.D, 1966). Based on the binomial model, the variance and mean plot of EPSCs is predicted to follow a parabola relationship which can be derived as follows:

MeanEPSC= N * Pr * q, and

VarianceEPSC =N * Pr * (1-Pr) * q²

If we combine the two equations above, we would obtain:

y = (1/N) * x² + q * x

where y: VarianceEPSC; x: MeanEPSC

Figure 4 illustrates how the parabola relationship of variance and mean develops with different Pr. When Pr is low, only the initial part of the parabola is plotted, so instead of a parabola, the V-M plot shows a linear relationship. When Pr is higher than 0.5, a parabola relationship can be observed.

Figure 4 Illustration of varianc- mean relationship

Fitting the data points of variance mean plot with a parabola, the q and N can be estimated. The initial slope of the parabola is q which can be derived as follows:

y‘ = (2/N) *x + q When x -> 0,

y‘ = q

From the intercept of the parabola with the x-axis N can be calculated.

y = 0, x = 0 or N * q Therefore,

N= x / q (when y = 0)

When N is known, the Pr can be further calculated by dividing the first EPSC by the product of N and q.

Pr = 1^st EPSC / (N * q)

Note that other sources of fluctuation make the analysis more complicated (see results) .

2.4 Deconvolution

The EPSCs were deconvolved with the miniature EPSC (mEPSC) to estimate transmitter release rates (Diamond and Jahr, 1995; Van der Kloot, 1988). In practice, we used the same procedure as Neher and Sakaba (2001), except that no residual current component due to delayed clearance of glutamate was assumed. In each cell pair, the

decay time constant of mEPSC was adjusted by varying the time constant until spontaneous events become delta-pulse like events in the release rate trace. The mEPSC amplitudes were assumed to be 5 pA in the presence of 0.5 mM Kyn, which was verified by variance-mean analysis (see below, Results).

2.5 Dynamic clamp experiments

The simulated excitatory synaptic current was injected to the soma by a home-built hardware realization of dynamic clamp setup. It composed of three stages: input signal filtering and conditioning (analog circuit), digitalization (Analog Devices AD 7495) and calculation (Atmel AT32AP7000), digital to analog conversion (Analog Devices DA5620). The control program was written in C (with development environment AVR32 Studio 2.5.0 and compiler avr32-gcc 4.3.2). The simulated synaptic current I was calculated from the voltage-independent conductance g and the instantaneous driving force, V – E.

I = g * (V - E)

V is the membrane potential measured from patch clamp amplifier and fed to the dynamic clamp in each cycle. Here the I-V relationship is assumed to be linear. E is the reversal potential for AMPA receptors, which is set to 0 mV. g is modeled by a double exponential waveform

g = G * ( - exp(-t/tau_rise) + exp(-t/tau_decay) ).

G is a scaling factor for the peak amplitude. Tau_rise and tau_decay were the time constants for the rising and decaying phase of the waveform and they are taken from fit to the recorded EPSCs. In each cycle of the dynamic clamp experiment, an updated I was calculated according to (1) and (2) and fed to the soma through the patch pipette. For each simulated synaptic event, the peak amplitude of the conductance was a positive random number from a Gaussian distribution with the calculated mean and S.D. from recorded EPSC peak amplitudes (Table 2). The update rate of dynamic clamp was 50 kHz. Experiments were done in the presence of 10-25µM bicuculine (Sigma).

Results

3.1 Variance-mean analysis and the mEPSC amplitude measurement at L4 neurons in V1

We performed simultaneous paired or triple recordings on the somata L4 RS neurons in mouse visual cortex. First, the neuronal type was identified according to the firing pattern of a neuron upon a current injection. These L4 neurons were then classified into three categories: RS neuron, FS neuron, and LTS neuron (Fig. 5) (Beierlein et al., 2003; Buhl et al., 1997; Gupta et al., 2000). While RS neurons form excitatory synapses onto the postsynaptic neuron, FS and LTS neurons form inhibitory synapses. To test whether a pair of neurons was synaptically connected, we depolarized one of the two neurons from -80mV to 0 mV for 2 ms to induce an action potential and examined if we could record mono-synaptic, time-locked responses from the other neuron, and vice versa.

Furthermore, to exclude di-synaptic connections, only those connections were selected for analysis in which the time difference between the onset of EPSCs and stimulation of the pre-synaptic neuron was restricted to less than 5ms. In this study, we focused on only RS-RS connections in L4.

Figure 5 Three classes of excitatory connections in L4 of V1

The configuration of a paired recording from two neurons under bright field, and fluorescence images are shown (left, and middle). The right panel shows the firing pattern of the post- synaptic neuron. (A) The connection between two RS neurons. The right panel shows the firing pattern of a RS neuron. (B) the connection between a RS neuron and a FS neuron. The right panel shows the firing pattern of a FS neuron. (C) the connection between a low-threshold spiking neuron and a LTS neuron. The right panel shows the firing pattern of a LTS neuron.

After a connected pair was identified, 50 Hz train stimulations (depolarizations from -80 mV to 0 mV for 2 ms, to induce an action potential at the soma, which spreads to the terminal) were applied to the presynaptic neuron and the EPSCs were recorded from the post-synaptic one. Figure 6A illustrates pre- and post-synaptic currents elicited by such stimulation under 2 mM extracellular Ca²⁺ at room temperature from a representative connection in V1. A train-stimulation allowed us to sample EPSCs at different releasing probabilities (Scheuss V. and Neher E., 2001) within one trace. By repeating this train stimulation with an interval of 10 s more than 20 times, variance and mean of the EPSC amplitudes of each stimulus were obtained, and variance-mean analysis was applied to

shows that the relationship between variance and mean was linear. This indicates that Pr is low, and only q could be obtained from the analysis. On average, the q estimated by variance-mean analysis was 9.2 ± 2.9 pA (n = 3, before correcting for the c.v. of the mEPSCs of 27%, see below).

In addition to variance-mean analysis, we looked into the individual traces of each connection. As shown in figure 6B, after the 9^th stimulus the EPSCs started to fluctuate in an all-or-none manner, and the amplitudes of the success events were the same, so it most likely resulted from a single vesicle release. On the contrary, the 5^th EPSC showed a variety of EPSC amplitude, indicating more than one vesicle was released from the terminal at the 5^th stimulus. We examined those EPSCs after the 9^th stimulus in those three connections, and 49 events were observed. The failure rate was higher than 0.5 at each stimulus (when we pooled all stimulus, the overall failure rate was higher than 0.9), and the average amplitude of successful events was 9.3 ± 0.7 pA (n=49). The amplitude here matched with the result of variance-mean analysis, implying that the estimates by our method were valid.

To further confirm our results, we measured the evoked mEPSCs and constructed the mEPSC histogram. For measuring the evoked mEPSCs, the extracellular Ca²⁺ was reduced to 0.4-0.6 mM (Isaacson and Walmsley, 1995; Katz and Miledi, 1965), and the divalent cation concentration was maintained by increasing the concentration of Mg²⁺ to 3 mM. A 2 ms depolarization (from -80mV to 0mV) was applied repetitively to the pre- synaptic cell with an interval of 3 s instead of 10-20 s in this particular set of experiments, and the EPSCs were recorded from the postsynaptic cell. We calculated the failure rate in each connection, and only those ones with a failure rate higher than 0.5 were taken for further analysis. Overall, 126 mEPSC events were recorded from 5 connections, and the average failure rate was 0.6. The average mEPSC amplitude was 10.8 ± 0.48 pA. The distribution of mEPSC amplitudes were plotted with that of train stimulation experiment in Figure 6C. The histogram of these two methods (evoked mEPSCs and EPSC events during the late period in the train) matched with each other quite well. Also the average

mEPSC amplitudes were similar and matched closely with quantal amplitudes estimated from variance-mean analysis.

Figure 6 Variance-mean analysis predicts quantal sizes, which matches with the mEPSC amplitudes

(A) The upper panel shows the presynaptic current of a representative RS-RS connection in V1 upon a short depolarization under room temperature, and the middle panel shows the mean EPSC corresponding to the stimuli from the same connection. The variance-mean relationship was plotted in the lower panel. Black solid circles represent the relationship of the variance and the mean of the EPSC amplitudes upon a stimulus train obtained from more than 20 repetitions.

The line fit (black line) estimates the q = 14.2 pA. (B) The individual EPSC traces from the same connection as (A) (middle panel). 10 individual traces out of 20 repetitions are superimposed to show the fluctuation of EPSCs. The grey traces are the close look at

obatiend under low external Ca²⁺(white bar). The inset shows representative mEPSC traces under low Ca²⁺. Grey bars indicate the histogram of mEPSCs obtained from train stimulation experiments.

The amplitude distribution of mEPSC measurement was slightly skewed to the right, and had a c.v. of 0.52. From the c.v. of mEPSC distribution, one can estimate the correction factor for q estimated from V-M analysis (see above). The correction factor was introduced here to estimate the maximum error caused by the dispersion of the mEPSC amplitudes, which causes fluctuation of EPSCs and is due to the variability with a given release site (intrasite variability) and among sites (intersite variability). The values of quantal size and binomial parameter N estimated from V-M analysis are denoted as q* and N*, respectively. These estimates can be subsequently corrected for the variability of mEPSC amplitude distributions, to give corrected quantal size q and binomial parameter N, according to Silver et al., 1998 as well as Scheuss and Neher, 2001:

q = q* / (1+c.v.²) and

N = N*  (1+W  c.v.²)

From Fig. 6C, one can calculate the c.v. of mEPSCs, which is 0.52. The symbol W represents the fraction of quantal variance that is caused by variability between different release sites (Frerking and Wilson, 1996). Because the source of variability in mEPSC amplitude distribution is not exactly known, we tentatively assumed W to be 0.5 (Meyer et al., 2001). Nevertheless, from c.v.=0.52, one can estimate 1+c.v.² = 1.27, suggesting that the actual q might be 27% smaller (Frerking and Wilson), and N can be 13% larger.

This error factor can be considered as an upper limit because the mEPSC histogram in Fig.6C may contain the release of more than two vesicles in some cases. If so, c.v. will be significantly smaller. Also, we have not corrected for the effect of jittering of release events (Taschenberger et al., 2005), which is known to cancel out the effect of the mEPSC dispersion. Therefore, the correction factor due to the mEPSC dispersion is likely to be the upper limit.

3.2 Estimation of N, Pr, and q under physiological condition with variance-mean analysis

To examine the quantal parameters under more physiological conditions, the recording temperature was raised up to 30-32 ℃ in the following experiments. We repeated the same 50Hz train stimulation protocol on RS-RS connections in V1, and applied variance-mean analysis to the evoked EPSCs in a stimulus train. In Figure 7A (left), the relationship of variance and mean could be fitted with a parabola in some connections. This indicates that temperature raises Pr significantly. From the parabola fit the q and N were estimated. The average q was 9.88 ± 1.09 pA (n = 5). Possibly, postsynaptic receptor saturation might have caused a parabolic relationship (Foster and Regehr, 2004; Meyer et al., 2001). Therefore, we applied 0.5mM Kyn to the brain slice, and the result is also shown in Figure 7A (black trace in the upper panel; black filled circles in the lower panel). Kyn is a low affinity AMPAR antagonist, so it competes with glutamate but does not block glutamate binding completely when the concentration was not too high. Therefore, it was used to prevent post-synaptic receptors from saturation and desensitization, and to measure the glutamate concentration at the synaptic cleft. An advanced application of this chemical is that one can detect the change in glutamate concentration by observing blocking efficiency under different conditions (Diamond and Jahr, 1997; Wadiche and Jahr, 2001). In this experiment, we mainly used it to protect the receptors from saturation and desensitization so that the assumption of contant quantal sizes in the V-M analysis holds true. With Kyn, the variance-mean relationship of EPSCs was linear in 3 connections (a representative connection is shown in Fig. 7A right) and was parabola in 4 connections (Fig. 7A left) out of 7 connections. Assuming the difference of these two groups (linear v.s. parabola) resulted from the different Pr, the N could be obtained only from those four with parabola relationship, and the Pr could be calculated only from them as well. The average N was 7 ± 1.1 (n = 4), and the average Pr of these 4 connections was 0.59 ± 0.05 (n = 4). However, the average Pr must be lower than 0.59 because of 3 linear cases (<< 0.5). This indicates that the release probability is heterogeneous at V1.

Though in some cases the Pr was high enough to show a parabolic relationship, the fraction of linear ones in the population was high. Therefore, we attempted to increase the Pr so that we could have more parabola cases, allowing one to estimate N in all the cases. To achieve this goal, we first tried the conventional way of increasing Pr by elevating the extracellular Ca²⁺ concentration to 4 or 8 mM Ca²⁺. However, we could not observe any augmentation of the evoked EPSCs upon high Ca²⁺ (data not shown). One possible reason could be the surface charge screening effect due to excess divalent cations. Hence, another method was tested, which was to apply 10 mM TEA, a K⁺ channel blocker, along with 0.5mM Kyn, a low affinity AMPAR antagonist, in the bath.

TEA is known to broaden the action potential waveform by blocking K⁺ channels and increase the Ca²⁺ flux such that we hoped to increase the Pr. Because rundown of release was faster under TEA (possibly due to the increased Pr), it was difficult to perform variance-mean analysis under both control and in the presence of TEA. Therefore in some experiments, only the TEA condition was tested. The result is shown in Figure 7B.

The average Pr was elevated to 0.69 ± 0.03 (n = 11), which showed that TEA did increase Pr. However, for a fraction of cells the linear relationship still remained (6 out of 17 connections). We pooled all parabola cases in both experiments (with Kyn only and with Kyn and TEA) assuming that the N should be intrinsically the same as that of other synapses, and the final estimate of N was 6.3 ± 1.0 (n = 13).

Figure 7 Estimation of N, Pr, and q in V1 RS-RS connections under physiological condition with variance-mean analysis

The upper panel shows the presynaptic currents of a representative RS-RS connection in V1 upon a short depolarization, and the middle panel shows the mean EPSCs in response to the stimuli from the same connection. The grey trace is the control group, and the black one was obtained under 0.5 mM Kyn. The lower panel shows the variance-mean relationship of the EPSCs both in control (black hollow circles) and in the presence of Kyn (black solid circle).

(left) the parabola fit of the control (black dash line) estimates N = 4.2 and q = 8.7 pA. The fit of the group under Kyn (black line) estimates N = 7 and q= 5 pA. (Right) The linear fit of the control (black dash line) estimates q = 7.8 pA. The fit of the group under Kyn (black line) estimates q = 3.3 pA .(B) The order of panels is the same as (A), but the traces are obtained from the experiments with TEA and Kyn in V1. The parabola fit estimates N = 7.6 and q = 3.9 pA.

3.3 Depletion of the readily releasable pool and calculating release rates with the deconvolution method

In order to verify the results from variance-mean analysis, we performed experiments to deplete the RRP of synaptic vesicles and measured the actual number of released vesicles. Part of the potassium in the intracellular solution was replaced with cesium, a K⁺ channel blocker, and the concentration ratio of Cs⁺ to K⁺ in the intracellular

such an extreme condition at physiological temperature. After a connected pair was identified, 10 mM TEA was applied to the bath to further block potassium channels and at the same time 0.5 mM Kyn was also added to prevent postsynaptic receptor saturation.

With both Cs⁺ and the external TEA, we aimed to have a step-like long depolarization at the terminal, similar to the idea of Katz and Miledi (1967) at the squid giant synapse. As shown in Figure 8 (upper panel), two 100 ms depolarizations were given to the presynaptic neuron with an interval of 200 ms, and the EPSCs were recorded from the postsynaptic one. The protocol was repeated in one connection several times with an interval of longer than 10 s for the cell to fully recover from the previous stimulus. The second depolarization was applied to confirm that the RRP was depleted completely, in which case there should be no response during the second pulse. That was the case in 7 cell pairs. While prefect voltage clamp of the presynaptic terminal was not expected, relatively short length of axons (Feldmeyer et al., 1999) together with blocking K⁺ channels with Cs⁺ might have helped sufficient stimulation of the terminal. The N was then calculated by the deconvolution method. Figure 8 (lower panel) shows the cumulative trace of release from a representative connection. On average, the N was 8.3

± 0.9 (n = 7). There was no significant difference between the results from variance- mean analysis and the depletion experiment (t-test, p = 0.13). Therefore, we concluded that the estimated N was valid (the assumption above was fulfilled). The result also suggests that a strong depolarization does not recruit additional pool of vesicles which are not used during an AP. The vesicle replenishment rate following the RRP depletion could be obtained from the cumulative trace. In V1, the time required to refill the whole RRP (Trec) under high [Ca²⁺] was 91.4 ± 15.5 ms (n = 7), which was estimated by fitting the cumulative release trace with an exponential (representing the RRP) and a line (representing the replenishment). One may expect to see more recovered response during the second pulse (applied at an 200 ms interval), given that replenishment rate is less than 100 ms. Possibly, the Ca²⁺-dependent component of vesicle replenishment is accelerated during the pulse, but drop in Ca²⁺ after the pulse would decrease the rate of replenishment (Dittman and Regehr, 1998; Hosoi et al., 2007; Wang and Kaczmarek, 1998).

Figure 8 Depletion of the readily releasable pool and calculating cumulative release with deconvolution method in S1 connections

A representative depletion trace from a representative RS-RS connection in V1. The stimulation protocol is shown in the upper panel. The presynaptic neuron was depolarized from -80 mV to 0 mV for 100 ms, and after an interval of 200 ms, the second 100 ms depolarization was applied to the cell to test the remaining vesicles within the RRP. One representative individual EPSC responding to a strong depolarization is shown in the middle panel. Upon the second depolarization, there was no vesicle released, indicating depletion of the RRP. The lower panel shows the cumulative trace of release from deconvolution of the EPSC. The two components (an exponential and a line) fit is shown in black dash line. From the fit, the number of vesicle in the RRP was estimated to be 7.4, the time constant of RRP depletion was 0.89 ms, and the vesicle replenishment rate constant was 70 ms in this case.

3.4 Saturation of postsynaptic receptors at RS-RS connections in S1

From the data above, we have obtained the three important quantal parameters (N, Pr, and q) of RS-RS synapses in V1. The N of RS-RS connections in V1 were quite different from those determined previously in S1synapses : According to Feldmeyer et al.

(1999), the morphological synaptic contact number was 3-4 (Egger et al., 1999).

Therefore, we examined whether the synaptic properties were different among different cortical areas. We repeated the variance-mean analysis and the pool depletion experiments at the RS-RS connections in the L4 of S1 at physiological temperature (30- 35℃). An example from the variance-mean analysis is shown in Figure 9 (lower panel).

In the control group, the variance exhibited a shallow dependence on the mean (black hollow circle). When a parabola was fitted to the data, the intercept of x-axis was far above 0 (black dash line). This indicates that the variance-mean analysis is not valid. It is possible that some post-synaptic factors, for example, saturation of the post-synaptic receptors (Foster and Regehr, 2004) distorted the relationship. To test this, we applied 0.5mM Kyn to the slice and as seen in Figure 9 (lower panel, black solid circle), the variance-mean relationship was restored to parabola shape. Therefore, at RS-RS synapses in S1 post-synaptic receptor saturation played a role during short-term synaptic plasticity, reducing synaptic depression (see below). Under Kyn, the N in S1 was 7.4 ± 1.3 and the Pr was 0.66 ± 0.03 (n = 9). There was no significant difference between both values in S1 and those in V1 when the data could be fitted with a parabola (t-test, for N, p

= 0.5; for Pr, p = 0.27). It is worthwhile to note that in S1 all connections showed a parabola relationship under Kyn whereas in V1 about a half of the population was linear.

Therefore, the overall Pr in V1 must be lower than S1 and more heterogeneous.

Figure 9 Estimation of N, Pr, and q q in S1 RS-RS connections under physiological condition with variance-mean analysis

The upper panel shows the presynaptic currents of a representative RS-RS connection in S1 upon a short depolarization, and the middle panel shows the mean EPSCs in response to the stimuli from the same connection. The grey trace is the control group, and the black one was obtained under 0.5 mM Kyn. The lower panel shows the variance-mean relationship of the EPSCs both in control (black hollow circles) and in the presence of Kyn (black solid circle).

The parabola fit of the group with Kyn (black filled circles) estimates N = 6.25 and q = 5.3 pA.

Next, we performed the RRP depletion experiment to confirm the N from variance- mean analysis. The N was 8.1 ± 2.0 (n = 7), which was not significantly different from the value from variance-mean analysis (t-test, p = 0.77). However, this number is much larger than previous estimates of the number of release sites at S1 (3 or 4, Feldmeyer et al., 1998; Buhl et al., 1997; see also Discussion). Trec at high [Ca²⁺] in S1 was 34.3 ± 6.4 ms (n = 7). The summary of N, Pr, q, and Trec of RS-RS synapses in both cortical areas is

Figure 10 Depletion of the readily releasable pool and calculating cumulative release with deconvolution method in S1 connections

A representative depletion trace from a representative RS-RS connection in S1. The stimulation protocol is shown in the upper panel. The presynaptic neuron was depolarized from -80 mV to 0 mV for 100 ms, and after an interval of 200 ms, the second 100 ms depolarization was applied to the cell to test the remaining vesicles within the RRP. One representative individual EPSC responding to a strong depolarization is shown in the middle panel. Upon the second depolarization, there was no vesicle released, indicating depletion of the RRP. The lower panel shows the cumulative trace of release from deconvolution of the EPSC. The two components (an exponential and a line) fit is shown in black dash line. The cumulative trace of release was filtered at 1 kHz, so it is less noisy than that in Fig.8, but this would not change the result. The fitting result shows the number of vesicles within the RRP was 7.8, the time constant of RRP depletion was 3 ms, and the replenishment rate constant was 28 ms in this example.

Q N Pr Trec

Variance-mean Mini Variance-mean Pool depletion Depression curve Pool depletion

V1

9.2±2.9 pA (n=3,RT)

9.0±0.9 pA (n=4,PT)

10.8±0.48 pA

7.0±1.1 (n=4, w/o TEA)

6.0 ±1.3 (n=11, w/ TEA)

8.3±0.9 (n=7)

<0.59±0.05^* (n=4, w/o TEA)

<0.69±0.03^* (n=11, w/ TEA)

100 ms (Pr = 0.45)

91.4±15.5 ms (n=7)

S1 7.4±1.3 (n=9)

8.1 ± 2 (n=7)

0.66 ± 0.03 (n=9)

71 ms (Pr =0.55)

34.3±6.4 ms (n=7)

* Considering in V1 some connections didn‘t show parabola relation in variance-mean analysis, the average Pr should be smaller than the value presented here in the table.

Table 1. Summary of the synaptic parameters under different conditions

Quantal sizes (q), the number of synaptic vesicles within the RRP (N) and release probability are shown. Q was obtained under control conditions, but variance-mean analysis did not work properly at S1. Under Kyn, qunatal sizes were estimated to be 5.5 ± 0.73 pA and 5.5 ± 0.7 pA in V1 and S1, respectively. Vesicle replenishment rates were estimated from the cumulative release trace in Fig. 8, 10 or else from the depression curve during a 50 Hz train under Kyn (Fig.

7, 9). A simple vescile pool depletion model (a single RRP with fixed replensiment rate) could explain the depression curve under kyn, when the Pr was set to 0.55 and 0.45 and the replenishment rate was set to 71 ms and 100 ms in S1 and V1, respectively.

In Fig. 10, the sustained EPSCs could be observed following depletion of the RRP.

While we assume that this component is mediated by synaptic vesicle replenishment, it is possible that the component is mediated, for example, by delayed clearance of glutamate in the synaptic cleft. In order to verify that it is indeed mediated by ongoing release events, fluctuation analysis was applied. Variance and mean was calculated during the sustained component, which should give a value of q by taking the ratio, multiplied by a factor of 2 (Katz and Miledi, 1972). On average, q of 5.9  0.6 pA was estimated at S1 (in the presence of 0.5 mM Kyn), which was close to the value estimated from variance- mean analysis used in Fig. 9. A close match between the two estimates suggests that the

fusion during vesicle replenishment, which means that MVR is mediated by coordinated fusion of synaptic vesicles in response to presynaptic Ca²⁺ influx. If so, then the quantal sizes would be increased. In V1, the analysis was more difficult to apply, possibly because of lower replenishment rates. Nevertheless, the q was estimated to be 7.8  0.7 pA.

3.5 A close look at the blocking effect of Kyn suggests MVR in S1.

As shown previously, Kyn restored the distorted variance-mean relationship of RS- RS connections in S1. Because differential block of Kyn is a strong indicator of MVR (Wadiche and Jahr, 2001), we looked into the blocking effect in more detail. Figure 11 illustrates the blocking efficiency of Kyn in both cortical areas. In the S1 group the second and subsequent EPSCs were blocked significantly more than the first one, whereas there is no significant difference in V1. Even in the connections with high Pr, there was no differential block at V1.

Figure 11 A close look at the blocking effect of 0.5 mM Kyn

The remaining fraction of EPSC after blocked by Kyn is shown in (A). The EPSCs amplitude of each peak with Kyn were divided by that of the control group. The figure shows the average ratio of the 1^st to the 4^th peak over all connections we recorded. The left panel shows the results from S1. There is significant difference between the 1^st peak and the rest (paired t-test, 1^st v.s.

2^nd: p = 0.0007; 1^st v.s. 3^rd: p = 0.0009; 1^st v.s. 4^th: p = 0.009). The right panel shows the results

from V1. There is no significant difference between the 1^st peak and the rest (paired t-test, 1^st v.s. 2^nd: p = 0.81; 1^st v.s. 3^rd: p = 0.72; 1^st v.s. 4^th: p = 0.9).

Differential block was not seen with NBQX (Fig. 12), a high affinity AMPAR antagonist. High affinity (or non competitive) antagonists should decrease the mEPSCs equally. If the differential block was still seen, the mechanisms other than MVR, most likley a voltage clamp problem must be involved (larger EPSCs have more clamp errors) (Wadiche and Jahr, 2001). Taken together, the results rather suggest that the local glutamate transient in the synaptic cleft changes during the train stimulation in active synapses in S1 but not in V1 under physiological condition.

Figure 12 The blocking effect of NBQX on the EPSCs at S1.

Similar to Fig 11, but a low concentration of NBQX (100 - 200 nM) was applied to the connected pair of L4 neurons in S1, to examine the blocking efficiency during a train stimulation. Because NBQX is a high affinity AMPA receptor antagonist, all the EPSCs would be blocked equally even in the presence of MVR. The Panel A shows a typical example